Andrew Wells on the Challenges of AI Without Formal Regulations

Hello, and welcome to the latest edition of the Techstrong AI video series. I'm your host, Mike Bor. Today we're with Andrew Wells, who's chief data and AI officer for NTT data, and we're talking about a report that they did that kind of shows that business people are, and tech leaders too, are a little uncertain about the future and the absence of AI regulations, and that's affecting their investment strategies.

Andrew, welcome the show. Great to be here, Mike. And, uh, very interesting what our survey found.

Yeah, I mean, well walk us through the high points, but you know, today we live in an era where everybody's running around saying, you know, we don't need regulations. And yet here we have business and IT leaders saying that in the absence of them, they're uncertain of essentially of how to proceed. How do we, this is a bit of a quandary as far as I can tell.

Yeah. You know, there's a dichotomy that exists with leaders that are wanting to drive a strong innovation agenda, but also wanna do it in a responsible manner. And that tug of war, you know, exists, you know, and that gap between those two, or it's probably wider than it's ever been.

The technology's changing so fast and you have this, you know, energy that's pushing you to wanna innovate and make sure that you continue your leadership position, or this is an opportunity to create a moat for your business, um, uh, by, uh, leveraging the technology change and doing a business process different. Uh, but, you know, you gotta do it in a responsible way as well. And it's that chasm, it's that, uh, it's that, uh, gap in between those two that I think a lot of the leaders you're struggling with.

What is it that you guys are recommending folks do to navigate that? I mean, do I just kind of follow the, uh, most stringent rules I can find on the assumption that everything else will be fine somewhere else? Or do I have different strategies for different geographies based on the regulations of the day?

Yeah, boy, I think it's very bespoke and probably one of the reasons why regulations are challenging in this space, because it really depends on the type of company you are. You know, if you're, um, manufacturing heavy or you got a heavy workforce, safety's a big issue. And, um, you've gotta really be fairly pedantic and, um, you know, concerned about the approach you take with the level of automation you drive with Gen AI and Genix to make sure people are kept safe, but you still need to lean in the change and solve for, well, how can this drive a better customer experience?

And how could I leverage this technology to do things, um, maybe faster or less costly or drive product? What most people are finding is it's driving productivity into the system to allow people to do more or do things faster. Do I kinda, as a business leader, am I worried about having to roll something back?

'cause uh, from once I roll it out and then I'm worried about some regulation that might show up six months from now and I don't have a lot of clarity into it, so does that make me a little more hesitant to invest in ai? Yeah, what we're finding is there's a lot of proof of concepts, but maybe one in 10, uh, are getting scaled into a production. So, and that's for a variety of reasons.

Some of them, it's, the technology's just not there yet, and there's not the efficacy around, um, the business case for rolling it out. But in some cases it's impact. It's what we are unsure of the impact or our workforce or our customers, uh, to be able to roll this out.

So we are going to, um, we're gonna wait and gauge it a little bit more, maybe continue to evolve it, but not scale it yet. Am I anxious because I'm worried that somebody else may not wait, and then I'm gonna have a competitive disadvantage. Because a lot of the times, uh, you know, you hear the phrase fomo, you're missing out.

Yeah, yeah. No, I definitely think there's a lot of that. I think maybe people are waiting for other people to move first before they roll something out or scale, although we're, we're not experiencing as much of that.

But I do think, uh, people are racing to drive the innovation, um, but they're also trying to gauge, well, how do I best drive the innovation responsibly? And there's, you know, we sort of think about it in four dimensions. One is leadership, and it needs to come from the top that as you innovate, you need to do it in a responsible manner.

And that we need to, um, carry both the, the ethics and the, um, the policies of our market and our organization into the work we do. I think second for us is, you know, designed by responsibility and the things that you design and things that you develop, make sure that responsible responsibility is inherent in those designs. Third for us would be governance.

You know, make sure you have multi-level governance in place, so if there are things that are being designed or, um, that you release, you can walk in the back fairly quickly. And then last is probably workforce readiness. How are you upskilling your workforce to really take advantage of these tools and understand how to use them in the right way?

It seems like to me too, we haven't quite fully solved this whole hallucination issue and, um, the models themselves are probabilistic, and that's a long fancy way of saying that they don't quite do the same thing the same way every time. Yeah. Have a lot of business processes that are deterministic where they have to be done the same way every time, so mm-hmm.

Is that a source of my, um, anxiety, because I'm not entirely sure what the outcome of these things is gonna be, even if it isn't malicious, it's gonna be different. Yeah, that's definitely a concern. You know, especially when you think about a agent and actions the models may take, um, as well as just analytics, you know, if you're looking to look at numbers that, um, you wanna make sure that the numbers that provides are accurate, it's, it's great right now for creativity, it's great.

Now, right now for writing, um, we do find a lot of efficacy in it around some of the analytics, but you need to, there's, there's methods and approaches for making sure that you can get tighter, um, answers back from your models, either through leveraging some rag techniques or through, uh, developing smaller models or training models, um, on your specific dataset. So it knows that, um, you know, it's got a tighter universe in terms of the answers it wants to provide, but it's, it's a challenge and you've gotta make sure that either if it's something that involves a, an important and impactful decision that a human's in the loop, uh, and or that you are looking at, well, what, what are the implications of this? And then how am I managing risk around it?

So you put in plenty of safeguards, but it, it is a concerned industry of where the models are now. The models are getting better though, and there are techniques you can, uh, deploy for driving accuracy. Is there anything in the report that kind of surprised you that you didn't think you'd see there?

Yeah, you know, I think the biggest one was the gap, uh, that we saw between, um, the two in terms of innovation and, uh, responsibility, uh, the, the, the tug of war of as well between executive leadership and then people that are implementing, uh, the technology sort of was another surprise. Executives are considerably more bullish on the technology then, um, maybe the people on the ground that are having to sort of stumble their way through it, because the technology is still very nascent. I think we find that most technology leaders or most leaders in organizations see the promise of the technology and what it could do to a particular process or a particular business, either, you know, creating a competitive differentiation for it, or driving revenue or saving cost.

But the people that are on the ground having to implement it are having to solve for, you know, a lot of the, um, uh, nascent stage of it. And there's a lot of bubble gum and, you know, uh, duct tape being used to sort of piece things together as, you know, uh, people put together the solutions. But just like any big wave in technology at the very beginning, it is, it's a little messy as, uh, the market innovates, but eventually it tightens up and, um, you know, you get solutions that are more scalable and, um, uh, drive a, uh, broader range of capabilities.

If I do start implementing all these AI capabilities and agents, it's not clear to me how much transparency do I have to provide to my partners and customers, uh, about those agents because they are, um, shall we say, potentially inconsistent at the very least. So do I need to disclose every time I'm using these AI agents or how will that play out? I do think there's gonna, especially around, uh, a decision that impacts something related to a customer, I think you're gonna find yourself needing to disclose the decision process.

And there are techniques that you can use to determine, you know what, maybe it's the weightings inside of a model, or maybe it's the reasoning behind the decision that you will, you know, I'm sure regulation's gonna come out and say, you eventually need to disclose those. So you need to build those into, into your solutions today to understand based off of the decision at this time, what did the model look like? Or what were the weights that were used or tell, have the model tell you, well, why, what was the reasoning path behind it?

So you can record those and make sure that you are, you know, being fair. And, um, you know, as you think about risk management, risk mitigation, you can look back and see, okay, well what was the model deciding? And then catch, even if it's in arrears where there may have been an issue, Ultimately what are the business leaders supposed to do?

I mean, am I quite literally gonna write my congressman and tell 'em to get their act together on regulations or, um, yeah. How will they make their voices heard? Gosh, I'd be surprised if anyone raises their hand and says, regulate me.

But I do think that, um, you know, if what we advise our clients to do is just make sure it's a principle responsible led approach, and that they are looking at how these decisions impact both their employees and their customers, and that they're putting in governance to that decision making process. So one, they can feel good that they're being responsible, and two, they can catch anything that is going on that may be because of a model hallucinating, or, you know, the fact that these technologies are new. Do you think that the employees that work on these projects will be the source of that kind of observation and information?

And maybe they, we all need to do a better job of listening to the folks in the front lines because, um, sometimes, you know, there's several layers removed from execution and the managers who signed this thing. So, um, will we need to have a a, a tighter loop of process re-engineering, I guess, between the employees at the front end and the execs at the back end? Yeah, it's gonna get really interesting, especially as, you know, gen AI starts to do a lot of the automation tasks.

You know, when it can, when you can tell Gen AI to go write a script for you and it does it, then how do you test it for its, um, you know, it being the script drives a responsible action. So, you know, my, my guess would be, and this is a guess, is that as these models get better at doing development, there's going to be models that are specifically created and reporting that's specifically created to tease out, well, where could some issues lie or where do the issues lie? So there will be safety checks that get put in place to make sure that you know, the code that's written, is it nefarious or doesn't drive a bad action?

But that's, it's a great challenge and it's gonna be one of those that I think the market's gonna solve for. Well, I did scratch my head about that very thought. 'cause in theory, I could use AI models to track the governance and the safety of the other AI models, but then how do I track the safety of this AI models that I put in place to make sure the other ones are safe?

Follow me? Yeah. That's sort of what Anthropics doing.

I mean, ROS niche in the marketplace is we're responsible AI and we're teaching our model to be more ethical than other models. And it's a really interesting niche that they're, they're heading down. So I do think it will exist.

I do think that, that it has a place in the marketplace. It has a space in the marketplace, and someone's gonna close it all. Right?

Yeah. You could get in an infinite loop of is my AI corrupting my AI to get the AI to do something nefarious. And, and if we got to that point, then We'll, we'll have different sets of problems then.

Yeah. Yeah. Hey, funks, you heard in here.

Here's a thought. If it feels bad, it probably is. So you should take another look at it before you go and implement this thing and have a conversation about it.

Andrew, thanks for being on the share. Thanks, Mike. All right.

And thank you all for watching The latest episode of the Textron AI video series can find this episode and others on our website. We invite you to check them all out. Until then, we'll see you next time.